MAEConformer for EEG and HRV Signal Representation Learning

MAEConformer uses a self-supervised masked autoencoder with Conformer architecture and multi-resolution short-time Fourier transform loss to learn from unlabeled physiological signals. The model was pretrained on over 6,000 hours of EEG and 4,800 hours of HRV recordings to capture multi-scale temporal and spectral features.